JP2009229191A - Quality prediction device, quality prediction method, quality prediction program and computer-readable storage medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a quality prediction device for easily and accurately predicting quality evaluation of food products at the equivalent level to that of a sensory test by an appraiser without depending on the appraiser. <P>SOLUTION: A quality prediction device 1 includes: a feature-amount extraction section 11 for extracting a feature amount of a spectrum of each food sample from results of an instrumental analysis on a plurality of food samples; a feature-amount-evaluation-value determination section 12 for determining a feature-amount evaluation value for evaluating the order of the feature amounts of the spectra classified according to wavelength; a correlation-degree calculation section 13 for determining a correlation degree between the feature-amount evaluation value and a quality evaluation value by a tester; a correlation-degree determination section 14 for determining whether the absolute value of the correlation degree is larger than a threshold value; an effective wavelength determination section 15 for determining the wavelength of the feature amount evaluation value used for determining the correlation degree by the correlation degree calculation section 13 as an effective wavelength when the absolute value of the correlation degree is determined larger than the threshold value; and a model formula formation section 16 for determining a quality prediction value of each food sample using the effective wavelength. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention makes it possible to easily and accurately predict the quality evaluation of food equivalent to the sensory test conducted by the appraiser without relying on the appraiser who appraise the quality of the food using the spectral data obtained by instrumental analysis. The present invention relates to a possible quality prediction apparatus, a quality prediction method, a quality prediction program, and a computer-readable recording medium.

  Conventionally, quality evaluations such as the taste and flavor of food depend on sensory tests. That is, the food quality evaluation is determined by a sensory test based on the appearance, aroma, color and taste of the food by an appraiser who assesses the quality of the food. However, in order to acquire the skills possessed by the appraiser, it took years of experience, and problems such as low objectivity and reproducibility occurred.

  In order to solve this problem, in recent years, a method for predicting the quality of food without depending on an appraiser has been studied. With this method, if all the components (chemical structure) contained in the food are extracted and the components of the extracted food can be quantified, it is possible to grasp the quality evaluation of the food thoroughly and accurately. Become. However, for example, in the case of green tea, the presence of about 600 compounds has been reported at present (Non-Patent Document 3), so it is difficult to predict the quality of food by a method of quantifying all the components. .

  Therefore, a method for predicting the quality of food by measuring the overall characteristics of the constituents of the food instead of quantifying all the constituents has been studied.

  As one of the measurement methods used for quantifying food constituents and predicting food quality, NIR (Near Infrared Spectrum; Near Infrared Spectroscopy) measurement methods such as FT-NIR (Fourier Transform Near Infrared) (Spectrum) measurement method. The FT-NIR measurement method is a method of continuously measuring the absorbance of a sample in the near-infrared wavelength region of 1000 nm to 2500 nm. By using the FT-NIR measurement method to quantify food constituents, the overall characteristics of food constituents (mixed ingredients) can be measured and used to predict food quality. It can be performed. That is, the FT-NIR measurement method is a method for measuring the overall characteristics of the food components, rather than separating and quantifying the food components.

  Here, as a method for predicting the quality of food using the FT-NIR measurement method, the applicant of the present invention has found an application according to Patent Document 1. In Patent Document 1, a process of performing an analysis sample by pre-processing a green tea sample, a process of obtaining an analysis result by performing instrument analysis on the analysis sample, and converting the analysis result into numerical data A green tea quality prediction method including a multivariate analysis process is disclosed. In addition, as an example of the quality prediction method for green tea, Patent Document 1 discloses that the quality of green tea can be simplified by performing multivariate analysis on a green tea sample subjected to instrument analysis using the FT-NIR measurement method. Methods that can be predicted are described. In the multivariate analysis in this case, the principal component analysis (PCA) is performed on the green tea sample to obtain a wave number region highly related to the grade of green tea, and then second order differentiation is performed on all data in this wave number region. ing. Non-patent document 2 also describes a quality prediction method for performing multivariate analysis on a green tea sample that has been instrumentally analyzed using the FT-NIR measurement method.

Furthermore, as one of multivariate analyzes using a method other than the PLS method, there is a moving window PLS (Moving Window-Partial Least Square; MW-PLS) method (Non-Patent Document 1). The moving window PLS (MW-PLS) method is a method capable of easily finding an effective wavelength and a wave number region and creating a calibration curve with high accuracy. In Non-Patent Document 1, the best spectral range of NH ₄ OH and H ₂ O ₂ is determined by using the MW-PLS method.

Non-Patent Document 3 discloses a method of evaluating characteristic aroma components contained in green tea using OASIS (Original Aroma Simultaneously Input to the Sniffing Port Method) method improved from AEDA (Aroma Extract Dilution Analysis) method. Yes. The OASIS method is a method in which a green tea volatile component separated by gas chromatography (GC) and an aroma generated from green tea are continuously mixed, and the characteristic aroma component contained in the green tea is evaluated by sniffing the mixed aroma. is there.
Japanese Patent Application No. 2008-20458 (filed on Jan. 31, 2008) Youngbok Lee, Hoeil Chung, Mark A. Arnold, “Improving the robustness of a partial least squares (PLS) model based on pure component selectivity analysis and range optimization: Case study for analysis of an etching solution containing hydrogen peroxide”, Analytica Chimica Acta 572 (2006) 93-101, 2006 Tatsuhiko Ikeda, Shigehiko Kanaya, Tsutomu Yonetani, Akio Kobayashi, and Eiichiro Fukusaki, “Prediction of Japanese Green Tea Ranking by Fourier Transform Near-Infrared Reflectance Spectroscopy”, J Agric Food Chem 55 (24): 9908-9912. (2007) Shoji Hattori, “Study on Green Tea Aroma Analysis Using the New Evaluation Method OASIS”, Memories of the Faculty of Agriculture, Hokkaido University, 28 (1): 85-120

  However, in the method described in Non-Patent Document 1, it is possible to search for an optimum region for obtaining a correlation formula, but it is not possible to simultaneously search a plurality of optimum regions that are separated from each other. Further, when there is a strong correlation between the absorbances between wavelengths in the moving window, the absorbances of all wavelengths in the moving window are used, which causes a problem that redundant correlation equations are obtained. .

  Patent Document 1 and Non-Patent Documents 1 to 3 disclose a method for determining a wavelength effective for quality prediction from sample data of green tea and performing the quality prediction of green tea using the determined effective wavelength. Not. In particular, in Patent Document 1 and Non-Patent Documents 1 to 3, a wavelength effective for quality prediction is determined from sample data of green tea after high-order differentiation is performed on the sample data of green tea, and the determined effective A method for predicting the quality of green tea using the wavelength is not disclosed.

  The present invention has been made in view of the above problems, and its purpose is the same as a sensory test conducted by an appraiser without relying on an appraiser who appraises the quality of food using spectral data obtained by instrumental analysis. It is an object of the present invention to provide a quality prediction apparatus capable of easily and accurately predicting quality evaluation of foods.

  The quality prediction apparatus according to the present invention is a quality prediction apparatus that performs food quality prediction in order to solve the above-described problem, and extracts spectral feature amounts for each food sample from the result of instrument analysis for a plurality of food samples. And a feature amount evaluation unit for classifying the feature amount of the spectrum extracted by the feature amount extraction unit for each food sample for each wavelength and evaluating the rank of the feature amount of the spectrum classified for each wavelength. A feature amount evaluation value determining means for determining a value, a correlation for obtaining a correlation degree between the feature amount evaluation value determined by the feature amount evaluation value determining means and a quality evaluation value by an expert who performs quality evaluation of the plurality of food samples Degree calculation means, correlation degree determination means for determining whether or not the absolute value of the correlation degree obtained by the correlation degree calculation means is greater than a preset correlation degree threshold, and the correlation degree determination means Therefore, when it is determined that the absolute value of the correlation degree is larger than the threshold value, effective wavelength determination is performed in which the wavelength in the feature amount evaluation value used when the correlation degree calculation unit obtains the correlation degree is determined as the effective wavelength. And a quality predicted value determining means for determining a quality predicted value of each food sample using the effective wavelength determined by the effective wavelength determining means.

  In order to solve the above problems, a quality prediction method according to the present invention is a quality prediction method in a quality prediction apparatus that performs quality prediction of food, and the feature amount of the spectrum is obtained from the equipment analysis results for a plurality of food samples. In order to classify the feature amount of the spectrum extracted for each sample in the feature amount extraction step to be extracted for each sample, and for each wavelength in the feature amount extraction step, and to evaluate the rank of the feature amount of the spectrum classified for each wavelength Correlation between the feature value evaluation value determining step for determining the feature value evaluation value, the feature value evaluation value determined in the feature value evaluation value determining step, and the quality evaluation value by the appraiser who performs quality evaluation of the plurality of food samples The correlation value calculating step for calculating the degree of correlation and the absolute value of the degree of correlation calculated in the above correlation degree calculating step is based on a preset correlation degree threshold value. And when the absolute value of the correlation is determined to be greater than the threshold value, the correlation is determined in the correlation calculation step. An effective wavelength determination step that determines the wavelength in the feature value evaluation value used as the effective wavelength, and a quality prediction value determination that determines the quality prediction value of each food sample using the effective wavelength determined in the effective wavelength determination step. And a step.

  According to the above configuration, in the quality prediction apparatus and the quality prediction method, the feature amount of the spectrum is extracted from the device analysis result, and the feature amount evaluation value is determined using the feature amount of the spectrum. The degree of correlation between this feature quantity evaluation value and the quality evaluation value by the appraiser is obtained. If the absolute value of this degree of correlation is greater than a preset threshold value, the feature quantity used when obtaining this correlation degree The wavelength in the evaluation value is determined as the effective wavelength. And the quality prediction value of each foodstuff is determined by using this effective wavelength. That is, in the quality prediction apparatus and the quality prediction method, only wavelengths that are appropriate for determining the quality predicted value of food are extracted from the wavelength range of the instrument analysis result, and the extracted wavelengths are determined as effective wavelengths.

  Thereby, in the quality prediction apparatus and the quality prediction method, the quality prediction of food can be performed with high accuracy, and the quality prediction result of food with high reliability can be provided. In other words, by using a quality prediction device and quality prediction method, it is possible to easily and accurately predict the quality evaluation of food equivalent to the sensory test conducted by the appraiser without relying on the appraiser who appraise the quality of the food. Is possible.

  Further, the quality prediction apparatus according to the present invention is characterized in that the feature quantity extraction unit obtains a higher-order differential coefficient for each wavelength by differentiating the device analysis result with respect to the wavelength, and the same among the obtained differential coefficients. A vector expression obtained by extracting the differential coefficient of the order may be extracted as the feature quantity of the spectrum.

  According to the above configuration, the feature quantity extraction unit extracts the differential coefficient of the same order and expresses it as a vector feature as the spectrum feature quantity. At this time, the feature quantity extraction means obtains a higher-order differential coefficient for each wavelength by differentiating the wavelength with respect to the device analysis result for this vector expression.

  Thereby, the effective wavelength determination means of the quality predicting apparatus can determine more wavelengths as effective wavelengths in the instrument analysis result. That is, in the quality prediction apparatus, a vector expression using a higher-order differential coefficient can be used as a spectral feature amount, and therefore, accurate quality prediction of food can be easily performed.

  Furthermore, in the quality prediction apparatus according to the present invention, the quality prediction value determination means may create a model formula for determining the quality prediction value of each food sample by performing multivariate analysis. In the quality prediction apparatus according to the present invention, the multivariate analysis may be regression analysis using a partial least square method.

  According to the above configuration, the quality prediction apparatus performs food quality prediction by creating (constructing) a model formula obtained by performing multivariate analysis. Moreover, in the quality prediction apparatus, regression analysis using a partial least square method is used as multivariate analysis.

  Thereby, the quality prediction apparatus can accurately predict the quality of food using the effective wavelength included in the instrument analysis result as a variable.

  Furthermore, in the quality prediction apparatus according to the present invention, the instrumental analysis result may be data indicating the absorbance of a food sample obtained by near infrared spectroscopy.

  According to the said structure, the quality prediction apparatus can perform the quality prediction of food accurately using the spectrum data which shows the light absorbency of a food sample.

  Furthermore, the quality prediction apparatus according to the present invention may further include display means for displaying an operation screen for acquiring a user operation.

  According to the above configuration, the user can easily cause the quality prediction apparatus to perform food quality prediction by operating the operation screen displayed on the display means, and can easily know the quality prediction result. it can.

  The quality prediction apparatus may be realized by a computer. In this case, a quality prediction program for causing the quality prediction apparatus to be realized by a computer by causing the computer to operate as the above-described means, and the program are recorded. Computer-readable recording media are also within the scope of the present invention.

  As described above, the quality prediction apparatus according to the present invention includes a feature amount extraction unit that extracts a spectrum feature amount for each food sample from the instrument analysis results for a plurality of food samples, and the feature amount extraction unit includes the food sample. A feature amount evaluation value determining means for classifying the feature amount of the spectrum extracted for each wavelength and determining a feature amount evaluation value for evaluating the rank of the feature amount of the spectrum classified for each wavelength; and the feature amount Correlation degree calculating means for obtaining a correlation degree between the feature amount evaluation value determined by the evaluation value determining means and a quality evaluation value by an appraiser who performs quality evaluation of the plurality of food samples, and a correlation degree obtained by the correlation degree calculating means Correlation degree determining means for determining whether or not the absolute value of the correlation degree is greater than a preset correlation degree threshold value, and the correlation degree determining means determines that the absolute value of the correlation degree is greater than the threshold value. The effective wavelength determining means for determining, as the effective wavelength, the wavelength in the feature amount evaluation value used when the correlation calculating means calculates the correlation, and the effective wavelength determined by the effective wavelength determining means. And a quality prediction value determining means for determining a quality prediction value of each food sample.

  In addition, as described above, the quality prediction method according to the present invention includes a feature amount extraction step for extracting a spectrum feature amount for each food sample from the device analysis results for a plurality of food samples, and the feature amount extraction step. A feature amount evaluation value determining step for classifying the feature amount of the spectrum extracted for each food sample for each wavelength, and determining a feature amount evaluation value for evaluating the rank of the feature amount of the spectrum classified for each wavelength, and Correlation degree calculation step for obtaining a correlation degree between the feature quantity evaluation value determined in the feature amount evaluation value determination step and a quality evaluation value by an appraiser who performs the quality evaluation of the plurality of food samples, and the correlation degree calculation step A correlation degree determination step for determining whether or not the absolute value of the correlation degree is larger than a preset correlation degree threshold; If the absolute value of the correlation degree is determined to be greater than the threshold value, the effective wavelength for determining the wavelength in the feature value evaluation value used when the correlation degree is obtained in the correlation degree calculation step as the effective wavelength. The method includes a determination step and a quality prediction value determination step of determining a quality prediction value of each food sample using the effective wavelength determined in the effective wavelength determination step.

  Therefore, the quality prediction apparatus and the quality prediction method can accurately predict the quality of food, and can provide a reliable food quality prediction result. In other words, by using a quality prediction device and quality prediction method, it is possible to easily and accurately predict the quality evaluation of food equivalent to the sensory test conducted by the appraiser without relying on the appraiser who appraise the quality of the food. Is possible.

  An embodiment of the present invention will be described below with reference to FIGS.

[Outline of Instrument Analyzer 2]
First, an instrument analyzer 2 that generates spectrum data to be supplied to the quality prediction apparatus 1 according to an embodiment of the present invention will be described. The instrumental analyzer 2 measures the absorbance of a food sample in an arbitrary wavelength region by an NIR (near infrared spectrum; near infrared spectroscopy) measurement method. For example, FT-NIR (Fourier transform near infrared spectrum) A measuring device is mentioned. The FT-NIR measurement device continuously measures the transmittance obtained in the wavelength range of 1000 nm to 2500 nm of the food sample, and converts this transmittance into absorbance. That is, it can be said that the FT-NIR measuring apparatus continuously measures the absorbance of the food sample in the wavelength range of 1000 nm to 2500 nm.

  Further, in the FT-NIR measuring apparatus, when the food whose quality is to be predicted is a solid such as green tea or black tea, the food is powdered, and then the powder is measured in a paste form with a solvent. As the solvent for making the powder into a paste, for example, a solvent that is usually selected according to the sample for adjusting the sample for IR measurement is used. In the FT-NIR measurement device, when the food whose quality is to be predicted is a liquid such as wine, the measurement is performed using the liquid as it is.

  As a result of instrument analysis measurement performed on the food sample by the instrument analyzer 2, spectrum data of the food sample as shown in FIG. 3 is obtained. The spectral data shown in FIG. 3 is spectral data when the food sample is green tea and the instrument analyzer 2 is an FT-NIR measuring device. That is, the instrument analyzer 2 is a measuring device for obtaining complex spectrum profile data of a food sample as shown in FIG.

  The quality prediction apparatus 1 according to the present invention predicts the quality of food using spectral data indicating absorbance obtained by the instrument analysis apparatus 2, and the details thereof will be described below. In the present embodiment, unless otherwise specified, the instrument analyzer 2 refers to an FT-NIR measuring device.

[Outline of Quality Prediction Device 1]
Here, a schematic configuration of the quality prediction apparatus 1 according to an embodiment of the present invention will be described with reference to FIG. FIG. 1 is a functional block diagram showing a schematic configuration of a quality prediction apparatus according to the present invention.

  The quality prediction apparatus 1 is an apparatus that can predict the quality evaluation equivalent to the sensory test performed by the appraiser without depending on the appraisers who appraise the quality of the food for the quality of various foods. That is, as will be described later, the quality prediction apparatus 1 determines a wavelength related to food quality prediction from the spectrum feature amount of the food sample, and constructs a model equation using the determined wavelength. Thereby, the quality prediction apparatus 1 estimates the quality predicted value of a food sample with high accuracy, that is, constructs (creates) a model formula having a high degree of correlation with the result of the sensory test (quality evaluation value).

  Here, as shown in FIG. 1, the quality prediction apparatus 1 according to the present invention includes a feature quantity extraction unit (feature quantity extraction means) 11, a feature quantity evaluation value determination section (feature quantity evaluation value determination means) 12, a correlation degree. Calculation unit (correlation degree calculation unit) 13, correlation degree determination unit (correlation degree determination unit) 14, effective wavelength determination unit (effective wavelength determination unit) 15, model formula creation unit (quality predicted value determination unit) 16, and display unit (Display means) 17 is provided as a functional block. Each of these functional blocks can be realized by, for example, a CPU (central processing unit) reading and executing a program stored in a ROM (read only memory) or the like into a RAM (random access memory) or the like.

  The feature quantity extraction unit 11 extracts a spectrum feature quantity from each spectrum data of a plurality of food samples acquired by the quality prediction apparatus 1 from the device analysis apparatus 2. Specifically, the feature amount extraction unit 11 performs the Taylor expansion at an arbitrary wavelength of each spectrum data, thereby extracting the feature amount of the spectrum as a form in which the differential coefficient of the same order is represented as a vector for each food sample. .

  The feature amount evaluation value determination unit 12 determines a feature amount evaluation value from the feature amount of the spectrum output from the feature amount extraction unit 11. Specifically, the feature quantity evaluation value determination unit 12 classifies the spectrum feature quantity extracted by the feature quantity extraction unit 11 for each food sample into differential coefficients for each wavelength, and then performs a wavelength-specific operation on these differential coefficients. Are ranked, and a feature value evaluation value (order of differential coefficients) for each wavelength is determined. That is, the feature amount evaluation value determination unit 12 determines a feature amount evaluation value for evaluating the rank of the feature amount of the spectrum classified for each wavelength.

  The degree-of-correlation calculation unit 13 uses the Spearman rank correlation coefficient as the degree of correlation between the feature amount evaluation value for each wavelength output from the feature amount evaluation value determination unit 12 and the quality evaluation value as a result of the appraisal. It is calculated using.

  The correlation level determination unit 14 compares the absolute value of the correlation level output from the correlation level calculation unit 13 with a preset threshold value and determines whether or not the absolute value of the correlation level is larger than the threshold value. It is. Further, when the correlation degree determination unit 14 determines that the absolute value of the correlation degree is larger than the threshold value, the correlation degree determination unit 14 outputs the wavelength in the feature amount evaluation value used when the correlation degree is calculated to the effective wavelength determination unit 15. .

  The effective wavelength determination unit 15 determines, as the effective wavelength, the wavelength in the feature amount evaluation value when the correlation degree determination unit 14 determines that the absolute value of the correlation degree is greater than the threshold value.

  The model formula creation unit 16 uses the effective wavelength determined by the effective wavelength determination unit 15 and the partial least square method (Partial Least Square method, PLS method) to determine the quality prediction value of each food sample. Is to create.

  The display unit 17 displays an operation screen, spectrum data acquired from the equipment analysis device 2, processing results in the quality prediction device 1, and the like. The display unit 17 displays, for example, a main screen 3 as an operation screen for acquiring user operations shown in FIG.

[Flow of processing in the quality prediction apparatus 1]
Next, the flow of processing in the quality prediction apparatus 1 according to the present invention will be described with reference to FIG. FIG. 2 is a flowchart showing the flow of processing in the quality prediction apparatus 1 according to the present invention. Here, the number of food samples whose quality is predicted by the quality prediction apparatus 1 is n.

  First, spectrum data of each food sample to be subjected to quality prediction in the quality prediction apparatus 1 is input as a measurement result of the instrument analysis apparatus 2 (S1). When the spectrum data is input to the quality prediction apparatus 1, the feature quantity extraction unit 11 extracts a spectrum feature quantity from the spectrum data for each food sample (S2).

  Specifically, when the feature quantity extraction unit 11 acquires the spectrum data of n food samples from the device analyzer 2, the feature quantity extraction unit 11 performs Taylor expansion for each arbitrary wavelength of the n spectrum data. That is, the feature quantity extraction unit 11 performs Taylor expansion represented by the following expression (1) around a wavelength b in the vicinity of an arbitrary wavelength s in the spectrum data received from the device analysis apparatus 2. That is, the feature quantity extraction unit 11 decomposes complex spectrum data for each wavelength up to higher-order differential coefficients.

  In the above description, the feature amount extraction unit 11 decomposes the complex spectrum data obtained by the instrument analyzer 2 into higher-order differential coefficients for each wavelength (that is, the higher-order differential coefficients are calculated from the spectrum data). For this purpose, the Taylor expansion method is used, but any method may be used as long as the same effect is obtained.

Then, the feature quantity extraction unit 11 _calculates a differential coefficient obtained by performing Taylor expansion in the vicinity of the wavelength s (s = s ₁ , s ₂ ,..., S _j ,..., S _m ). By expressing as shown in the following equation (2), the feature amount of the spectrum is extracted. Here, the equation (2) is the k-th order (same order) differentiation when the Taylor expansion is performed up to the k-th order differentiation for the spectrum data in the food sample i (i = 1, 2,..., N). It shows the feature quantity x _i ^(k) of the spectrum obtained as a form in which the coefficient is expressed as a vector. The wavelength _{s 1, s 2, ···,} s m shows the arbitrary wavelength spectral data of food samples i.

Next, the feature extraction unit 11, the spectrum of the characteristic amounts x i (k) When extracted for each n-number of food samples, the feature of these spectra x i (k) is the feature amount evaluation value determination section 12 Output. Feature amount evaluation value determining unit 12 may classify features x i of spectrum output from the feature extractor 11 (k) wavelength s 1, s 2, ···, each s m (S3), each wavelength A feature amount evaluation value using the feature amount of the spectrum classified into (1) is determined (S4).

Specifically, for example, in the case of the wavelength s ₁ , the feature amount evaluation value determination unit 12 determines the differential coefficient (f ₁ ^(k) ( ⁾ from the spectral feature amount x _i ^{(k) of} the n food samples to the wavelength s _1. s ₁ ), f ₂ ^(k) (s ₁ ),..., f _n ^(k) (s ₁ )) are extracted. That is, the feature value evaluation value determination unit 12 extracts the differential coefficient for each wavelength s _j (j = 1,..., ^M) from the spectrum feature value x _i ^(k), thereby obtaining m wavelengths. Classify into derivative.

When the feature quantity evaluation value determination unit 12 classifies the spectrum feature quantity x _i ^(k) into differential coefficients for m wavelengths, the feature quantity evaluation value determination unit 12 ranks the differential coefficients for each wavelength, and the feature quantity evaluation value ( Derivative coefficient order) xrank is determined. That is, the feature quantity evaluation value determination unit 12 converts the differential value that is a real value into an integer value by determining the feature quantity evaluation value xrank. Note that the feature value evaluation value at the wavelength s _j (k-th derivative) of the food sample i is expressed as xrank _i ^(k) (s _j ).

For example, in the case of differential coefficients (f ₁ ^(k) (s ₁ ), f ₂ ^(k) (s ₁ ),..., F _n ^(k) (s ₁ )) of the wavelength s ₁ , f ₂ ^{(k )} Consider the case where (s ₁ ) <f ₁ ^(k) (s ₁ ) <... <F _n ^(k) (s ₁ ). In this case, the feature value evaluation value xrank ₁ ^(k) (s ₁ ) = 2 at the wavelength s ₁ and xrank ₂ ^(k) (s ₁ ) = 1. That is, the feature quantity evaluation value determination unit 12 ranks the differential coefficients by rearranging the differential coefficients in ascending order of the real values.

  Next, the feature amount evaluation value determination unit 12 outputs the feature amount evaluation value for each wavelength to the correlation calculation unit 13. The correlation degree calculation unit 13 calculates the degree of correlation between the feature amount evaluation value xrank for each wavelength output from the feature amount evaluation value determination unit 12 and the quality evaluation value Y that is the result of appraisal by the appraiser (S5). Specifically, the correlation calculation unit 13 uses the Spearman rank correlation coefficient expressed by the following equation (3) to obtain the correlation r between the feature value evaluation value xrank and the quality evaluation value Y for each wavelength. calculate.

  In addition, the quality prediction apparatus 1 may acquire the quality evaluation value Y of the food sample measured by the instrument analyzer 2 together with the spectrum data from the instrument analyzer 2, or may be acquired by user input. The correlation calculation unit 13 uses Spearman's rank correlation coefficient in calculating the correlation r. However, the present invention is not limited to this, and any method can be used as long as the correlation r can be calculated. May be.

The correlation degree calculation unit 13 outputs the calculated correlation degree r to the correlation degree determination unit 14. The correlation degree determination unit 14 compares the absolute value of the correlation degree r output from the correlation degree calculation unit 13 with a preset threshold value r _th, and determines whether the absolute value of the correlation degree r is greater than the threshold value r _th . It is determined whether or not (S6). That is, the correlation degree determination unit 14

It is determined whether or not. The threshold value r _th is set to 0.90, for example. However, the present invention is not limited to this, and it is only necessary to set a threshold value that normally determines that there is a correlation with the correlation degree r. Further, the threshold r _th may be appropriately set by the user when using the quality prediction apparatus 1 or may be set in advance when the quality prediction apparatus 1 is manufactured.

If the correlation degree determination unit 14 determines that the correlation degree r is greater than the threshold value r _th (YES in S6), the correlation s in the feature amount evaluation value xrank used when the correlation degree r is calculated is valid. Output to the wavelength determination unit 15. That is, the correlation degree determination unit 14 determines that the correlation between the feature amount evaluation value xrank determined by the feature amount evaluation value determination unit 12 and the quality evaluation value Y is high when YES in S6. On the other hand, when the correlation degree determination unit 14 determines that the correlation degree r is equal to or less than the threshold value r _th (NO in S6), the process in the quality prediction apparatus 1 is terminated.

Upon receiving the wavelength s output from the correlation degree determination unit 14, the effective wavelength determination unit 15 determines that this wavelength s is effective as a wavelength used in the model formula y created by the model formula creation unit 16, and this wavelength s effective wavelength ^{s' -} determining a (S7). That is, the effective wavelength determining unit 15 determines the wavelength s in the feature amount evaluation value xrank when the correlation degree determining unit 14 determines that the correlation degree r is greater than the threshold value r _th as the effective wavelength s ^′ . Then, the effective wavelength determining unit 15 outputs the determined effective wavelength s ^′ to the model formula creating unit 16.

The model formula creation unit 16 uses the effective wavelength s ^′ output from the effective wavelength determination unit 15 and the PLS method to obtain a linear model formula y that determines the quality prediction value of each food sample by the following formula ( 4) (S8).

S ^′ ₁ ,..., S ^′ _j ,..., S ^′ _m in the equation (4) are wavelengths determined as the effective wavelength s ^′ by the effective wavelength determination unit 15.

The model formula creation unit 16 determines the number of factors using the prediction error (prediction capability) R _pred ² when creating the model formula y shown in formula (4). The PLS method and the prediction error R _pred ² used when the model formula creating unit 16 creates the model formula y will be described later.

  As described above, the quality prediction apparatus 1 can create the model formula y having a high degree of correlation with the quality evaluation value Y by the appraiser by creating the model formula y as described above. Therefore, the quality prediction apparatus 1 according to the present invention can accurately predict the quality of food, and can provide a highly reliable food quality prediction result. That is, by using the quality prediction device 1, it is possible to easily and accurately predict the quality evaluation of food equivalent to the sensory test performed by the appraiser without depending on the appraiser who appraises the quality of the food. Become.

[Create model formula by PLS method]
The PLS method used in the model formula creation unit 16 will be described. The PLS method is a statistical method called regression analysis used in multivariate analysis, and is an effective method for creating a calibration curve from spectrum data having a correlation between variables (for example, wave number and wavelength). Multivariate analysis is generally a statistical technique for analyzing the correlation between variables based on data on a plurality of variables. Thereby, the quality prediction apparatus 1 can perform quality prediction of food using the effective wavelength contained in the spectrum data output from the equipment analysis apparatus 2 as a variable.

  In general, if the correlation between variables is high, the regression accuracy will be significantly reduced depending on the combination of variables used. However, in order to avoid this, the PLS method converts the variables into mutually uncorrelated variables (latent variables). Perform regression using latent variables. That is, the PLS method is an analysis method in which data variables are orthogonally transformed and (multiple) regression analysis is performed using the new variables.

More specifically, the PLS method will be described. Here, in order to simplify the description, the k-th order differential coefficient f _i ^(k) (S ^′ _j ) of the effective wavelength s ^′ _j in the food sample i is expressed as x _ij . Further, the number of food samples is N, and the number of effective wavelengths is M. Note that the model equation y _i and the k-th order differential coefficient x _ij are

And converted. That is, here, as the model expression y by the PLS method,

Will be asked.

  First, if the feature amount and model expression of all spectra for N food samples are described in a matrix, the feature amount X and model equation y of all spectra are

It can be expressed. The PLS method uses Lagrange's undetermined multiplier method to derive equation (5).

  That is, w is defined as a vector of size 1 consisting of M elements, and the undetermined multiplier is μ,

far. Here, G (w, μ) is partially differentiated by w _k , and G (w, μ) _obtained by partial differentiation is set to 0.

It becomes. Solving this, w _k becomes

Is required. Next, using w _k (k = 1), the first latent variable t ₁

Then, using this first latent variable t ₁ ,

Ask for. Here, p ₁ is a loading vector of the first component, and q ₁ is a coefficient of the first component. From the above, the first parameters (t ₁ , p ₁ , q ₁ ) are obtained.

Next, when p ₁ and q ₁ are solved,

It can be expressed as. E is a matrix indicating the residual, and f is a vector indicating the residual. continue,

And the second parameter (t ₂ , p ₂ , q ₂ ) is obtained by repeating the processes after the equation (6). By repeating this A times, A set of parameters (t ₁ , p ₁ , q ₁ ), (t ₂ , p ₂ , q ₂ ),..., (T _A , p _A , q _A ) are obtained. be able to.

In addition, when p _k (k = 1,..., A) and q _k (k = 1,..., A) are solved using A set of parameters,

It can be expressed as. Here, T, P, and q in the above formula are

It can be expressed as. Here, T, P, and q are latent variables, loading vectors, and coefficients each consisting of A factors.

Further, the coefficient a _k (k = 1,..., A) in the equation (5) is

It can be obtained more. Here, W in the above equation is a weight vector composed of A factors expressed as W = [w ₁ , w ₂ ,..., W _A ]. The coefficient a ₀ is

And

Next, in order to determine the optimum number of factors (the number of A; the number of latent variables) in the model formula y, here, the prediction error R _pred ² (A) is obtained using a verification method called cross-validation. Cross-validation is a method in which measurement data (here, spectrum data measured by the instrument analyzer 2) is divided into estimation and verification, and the number of factors is determined so that the prediction error at the time of verification is minimized. .

The number of components of the remaining food sample excluding the u-th food sample out of N food samples (number of effective wavelengths s ^′ ) A = 1, 2,..., Model formula (the above formula (5) ) At this time it sought predicted value of each _{y u (1), y u} (2), ···, y u (M), the measured value and _y u (obs).

  The same operation is repeated M times for all food samples. And

And A when maximizing the prediction error R _pred ² (A) is determined as the optimum factor number. That is, the number of factors refers to the number of factors A included in the weight vector W, and the optimum weight vector W = [w ₁ , w _{2 is} used by using all the effective wavelengths s ^′ determined by the effective wavelength determining unit 15. ,..., W _A ].

〔Example〕
Next, an example of food quality prediction in the quality prediction apparatus 1 according to the present invention will be described below. In this embodiment, the quality prediction apparatus 1 is made to perform quality prediction using 13 samples of green tea (produced in Nara Prefecture: Yamato tea) that has been evaluated for quality by 1 to 64 ranks by an appraiser (Meister). It is.

  First, dry tea leaves (200 mg) are put into an Eppendorf tube (2 ml), pulverized at 20 Hz for 10 minutes using a ball mill, glycerol (600 μl) is further added, and the pulverized dry tea leaves are dissolved in glycerol. Then, the dried tea leaves dissolved in glycerol are homogenized again at 20 Hz for 10 minutes using a ball mill to produce a pasty green tea sample. The solution for dissolving the dried tea leaves is not limited to glycerol, but may be hexane, acetonitrile, perfume oil, or the like.

Next, using the FT-NIR measuring apparatus as the instrument analyzer 2, the absorbance of the obtained paste-like sample was measured in the wavelength range of 1000 nm to 2500 nm. In this example, as an FT-NIR measurement apparatus, a NEAR TYPE equipped with a Smart Near-IR UpDRIFT with a mirror speed of 1.2659 cm / s and a resolution of 8 cm ⁻¹ , a CaF ₂ beam splitter, and a cooled InGaAs detector. 6700 FT-IR (Thermo Electron Co., Ltd.) was used. The total number of data points was 1557 (1000 nm to 2500 nm) for each spectrum data.

As described above, the feature quantity extraction unit 11 of the quality prediction apparatus 1 acquires the spectrum data indicating the absorbance of the green tea sample from the instrument analysis apparatus 2, and performs 0 to 4 by performing Taylor expansion on the spectrum data. The following differential coefficients are obtained, and these differential coefficients are expressed as spectral feature amounts x _i ^(k) (k ≦ 4).

Next, the feature quantity evaluation value determination unit 12 classifies the spectrum feature quantity x _i ^(k) obtained by the feature quantity extraction unit 11 for each wavelength, and uses the classified spectrum feature quantity to evaluate the feature quantity evaluation value. xrank is determined and output to the correlation calculation unit 13. The degree-of-correlation calculation unit 13 uses the Spearman rank correlation coefficient to calculate the degree of correlation r between the feature amount evaluation value xrank determined by the feature amount evaluation value determination unit 12 and the quality evaluation value Y that is the result of the appraisal by the appraiser. Is calculated.

When the correlation level determination unit 14 determines that the correlation level r calculated by the correlation level calculation unit 13 is larger than the threshold value r _th = 0.90, the feature amount evaluation used when the correlation level r is calculated The wavelength s at the value xrank is output to the effective wavelength determining unit 15. Then, the effective wavelength determination unit 15 determines the wavelength s output from the correlation determination unit 14 as the effective wavelength s ^′ .

  Here, FIG. 4 is a diagram showing the spectral data indicating the absorbance of a certain green tea sample and the degree of correlation r obtained in the quality prediction apparatus 1. 4A is spectral data (raw data) indicating the absorbance of the green tea sample obtained from the instrument analyzer 2, and FIGS. 4B to 4E are the spectral data (raw data), respectively. It is the spectrum data when performing the 1st-4th order differentiation with respect to. In the graph shown in FIG. 4, the vertical axis represents the correlation degree r and the horizontal axis represents the wavelength (nm).

  Furthermore, “the number of effective measurement points (| r |> 0.9)” illustrated in FIG. 4 indicates the number of effective wavelengths determined by the effective wavelength determination unit 15, and (E) is shown corresponding to each.

As shown in FIG. 4A, in the case of spectrum data (raw data) (in the case of zero-order differentiation), the threshold r _th = 0. The number of wavelengths determined to be greater than 90, that is, the number of wavelengths determined by the effective wavelength determination unit 15 as the effective wavelength s ^′ was 0. Further, as shown in (b), the number of wavelengths determined as the effective wavelength s ^′ by the correlation degree determination unit 14 was also zero in the case of the first derivative.

Further, as shown in FIGS. 4C to 4E, the number of wavelengths determined by the effective wavelength determining unit 15 as the effective wavelength s ^′ is 10 in the case of the second derivative and in the case of the third derivative. In the case of the fourth-order differentiation, the number is 22 and the number is 20.

In other words, the effective wavelength determining unit 15 determines the effective wavelength s ^′ if the feature value x _i ^{(k) of the} spectrum extracted by the feature value extracting unit 11 is a differential coefficient of second order (k = 2) or more. can do. Also, the effective wavelength determining unit 15 can determine more wavelengths as effective wavelengths s ^′ in the spectrum data when the spectral feature amount x _i ^(k) is the third order (k = 3) or more. .

Next, the effective wavelength determination unit 15 determines the wavelength s when the correlation degree r is determined to be greater than the threshold value r _th (= 0.90) by the correlation degree determination unit 14 as the effective wavelength s ^′. The effective wavelength s ^′ is output to the model formula creation unit 16. The model formula creation unit 16 creates a model formula y using the effective wavelength s ^′ and the PLS method.

Here, FIG. 5 is a diagram showing evaluation results of prediction errors in a PLS regression equation (model equation) created by using the PLS method for the second to fourth order differential coefficients. In the graph shown in FIG. 5, the vertical axis represents the prediction error R _pred ² , and the horizontal axis represents the number of latent variables (number of factors). Moreover, (a)-(c) of FIG. 6 is a figure which shows the coefficient of the PLS regression equation in each effective wavelength in a 2nd-4th order differential coefficient, respectively. In FIG. 6, the contribution to the PLS regression equation for each wavelength can be grasped.

  Based on the prediction error evaluation results shown in FIG. 5, the model formula creation unit 16 has four factors and third-order differential coefficients in the PLS regression formula created in the case of the second derivative (2nd derivative). Each PLS regression equation was constructed by determining 3 factors in the case of (3rd derivative) and 5 factors in the case of 4th derivative. Since the PLS method and the prediction error have been described above, description thereof is omitted here.

  FIG. 7 is a diagram showing the degree of correlation between the predicted quality of green tea indicated by the model formula created by the model formula creation unit 16 and the quality evaluation value of green tea that is the result of appraisal by the appraiser. In the graph shown in FIG. 7, the vertical axis indicates the quality predicted value (Predicated) indicated by the model formula y, and the vertical axis indicates the quality evaluation value (observed) that is the result of the appraisal performed by the appraiser. That is, when the quality prediction value of green tea in the quality prediction apparatus 1 and the quality evaluation value by the appraiser match, each symbol shown in FIG. 7 (in the case of secondary differentiation; in the case of quadratic differentiation, in the case of cubic differentiation; triangle, 4 In the case of the second derivative; a circle) is plotted on the alternate long and short dash line shown in FIG.

  As shown in FIG. 7, the quality prediction value indicated by the model formula y created in the case of the second to fourth order differential coefficients is the degree of correlation (correlation degree) with the quality evaluation value by the expert in any of the 13 green tea samples. 0.95 or higher) is high.

  That is, in the quality prediction apparatus 1, even when green tea is used as the food, the model formula creating unit 16 creates the model formula y as described above, so that the model having a high degree of correlation with the quality evaluation value by the appraiser. Expression y can be created. Therefore, the quality prediction apparatus 1 according to the present invention can accurately predict the quality of green tea, and can provide a reliable quality prediction result of green tea.

  In the above description, the quality prediction apparatus 1 uses green tea as a food for which quality prediction is desired. However, the present invention is not limited to this, and the same applies to foods that are often evaluated by ordinary appraisers such as tea and wine. The effect of can be obtained.

  Here, a quality prediction program capable of executing a series of processes performed by the quality prediction apparatus 1 and the equipment analysis apparatus 2 in the present embodiment is created using the Java (registered trademark) language. The language used for the quality prediction program is not limited to the Java (registered trademark) language, and may be any programming language that can create a program for a series of processes performed by the quality prediction apparatus 1 and the equipment analysis apparatus 2. That's fine. In addition, the quality prediction program should just be a program which can perform the process performed with the quality prediction apparatus 1 at least.

  When the quality prediction device 1 and the device analysis device 2 detect a user operation with respect to various buttons displayed on the main screen (operation screen) 3 shown in FIG. 8, the quality prediction device 1 and the device analysis device 2 perform processing according to the function of the various buttons. . Here, FIG. 8 is a diagram illustrating a main screen that is activated when processing in the quality prediction apparatus 1 and the equipment analysis apparatus 2 is executed, and an example of a screen displayed on the display unit 17 of the quality prediction apparatus 1. It is. The main screen 3 may not be displayed on the display unit 17 of the quality prediction apparatus 1, and may be displayed by a display means provided separately from the quality prediction apparatus 1. Hereinafter, functions of various buttons displayed on the main screen 3 shown in FIG. 8 will be described.

The main screen 3 shown in FIG. 8, "File Selection (1 / cm)" button 31, "Differentiation" button 32, "Spearman ^'s cc" button 33, "Peak Selection" button 34, "PLS" button 35, An “Estimation by PLS model” button 36 and a “PLS (cross-validation)” button 37 are included.

  The “File Selection (1 / cm)” button 31 is used to cause the instrument analyzer 2 to execute processing, and specifically, to convert the transmittance obtained in the wavelength range of the food sample from 1000 nm to 2500 nm into absorbance. It is.

  The “Differentiation” button 32 causes the quality prediction apparatus 1 to execute processing. Specifically, the “Differentiation” button 32 performs high-order differentiation on the spectrum data that is the measurement result of the instrument analyzer 2, and performs high-order differentiation. This is for obtaining a differential coefficient. Although commercially available software can only obtain differential coefficients up to the second order, the quality prediction program used in the quality prediction apparatus 1 according to the present invention can obtain differential coefficients up to the fourth order. is there.

"Spearman ^'s cc" button 33 is intended to execute the process to quality prediction apparatus 1, specifically, those that cause seeking Spearman's rank correlation. The “Peak Selection” button 34 is for causing the quality prediction apparatus 1 to execute processing, and specifically, for determining a threshold value according to user settings.

  The “PLS” button 35 is used to cause the quality prediction apparatus 1 to execute processing, and specifically, to obtain a PLS regression equation. The “Estimation by PLS model” button 36 is used to cause the quality prediction apparatus 1 to execute processing. Specifically, the quality prediction of the food by the PLS regression equation is performed (that is, the food quality ranking (quality Evaluation). Further, the “PLS (cross-validation)” button 37 is for causing the quality prediction apparatus 1 to execute processing, and specifically, for determining the number of effective factors of the PLS regression equation.

  FIG. 9 is a diagram illustrating an example of a model formula created when processing in the quality prediction apparatus 1 is performed on a green tea sample. FIG. 9 shows a model formula y created when the model formula creating unit 16 obtains a third order differential coefficient using the green tea sample obtained when the spectrum data shown in FIG. 4 is obtained.

In FIG. 9, the model formula creation unit 16 uses all the effective wavelengths s ^′ determined by the effective wavelength determination unit 15 (22 in the case of a third-order differential coefficient; see FIG. 4) to obtain the model formula y. Indicates the case of creation. The model formula creating unit 16 uses the number of factors when the prediction error R _pred ² shown in FIG. 5 becomes maximum (three factors in the case of the third order differential coefficient) from the model formula y shown in FIG. It is also possible to create a model equation y.

  As described above, when the quality prediction device 1 and the device analysis device 2 detect a user operation on various buttons on the main screen 3 illustrated in FIG. 8, the quality prediction device 1 and the device analysis device 2 perform the above-described processing according to the functions of the various buttons. Accordingly, the user can easily cause the quality prediction apparatus 1 to perform food quality prediction by operating various buttons on the main screen 3, and can easily know the quality prediction result.

[Supplement]
Finally, each block of the quality prediction apparatus 1 according to the present embodiment, in particular, the feature amount extraction unit 11, the feature amount evaluation value determination unit 12, the correlation degree calculation unit 13, the correlation degree determination unit 14, the effective wavelength determination unit 15, and The model formula creation unit 16 may be configured by hardware logic, or may be realized by software using a CPU as follows.

  That is, the quality prediction apparatus 1 according to the present embodiment includes a central processing unit (CPU) that executes instructions of a control program that realizes each function, a read only memory (ROM) that stores the program, and a RAM that expands the program. (Random access memory), a storage device (recording medium) such as a memory for storing the program and various data. An object of the present invention is a recording medium on which a program code (execution format program, intermediate code program, source program) of a control program of the quality prediction apparatus 1 which is software that realizes the above-described functions is recorded so as to be readable by a computer. This can also be achieved by supplying the quality prediction apparatus 1 and reading and executing the program code recorded on the recording medium by the computer (or CPU or MPU).

  Examples of the recording medium include a tape system such as a magnetic tape and a cassette tape, a magnetic disk such as a floppy (registered trademark) disk / hard disk, and an optical disk such as a CD-ROM / MO / MD / DVD / CD-R. Card system such as IC card, IC card (including memory card) / optical card, or semiconductor memory system such as mask ROM / EPROM / EEPROM / flash ROM.

  The quality prediction apparatus 1 according to the present embodiment may be configured to be connectable to a communication network, and the program code may be supplied via the communication network. The communication network is not particularly limited. For example, the Internet, intranet, extranet, LAN, ISDN, VAN, CATV communication network, virtual private network, telephone line network, mobile communication network, satellite communication. A net or the like is available. Also, the transmission medium constituting the communication network is not particularly limited. For example, even in the case of wired such as IEEE 1394, USB, power line carrier, cable TV line, telephone line, ADSL line, etc., infrared rays such as IrDA and remote control, Bluetooth ( (Registered trademark), 802.11 wireless, HDR, mobile phone network, satellite line, terrestrial digital network, and the like can also be used. The present invention can also be realized in the form of a computer data signal embedded in a carrier wave in which the program code is embodied by electronic transmission.

  The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope shown in the claims. That is, embodiments obtained by combining technical means appropriately modified within the scope of the claims are also included in the technical scope of the present invention.

  The quality prediction apparatus according to the present invention can easily and accurately predict the quality evaluation of food equivalent to the sensory test conducted by the appraiser without relying on the appraiser who appraise the quality of the food. For example, it is effective for foods often requiring quality evaluation by an expert such as green tea, black tea, and wine.

It is a functional block diagram which shows schematic structure of the quality prediction apparatus which concerns on one Embodiment of this invention. It is a flowchart which shows the flow of a process in the quality prediction apparatus shown in FIG. It is a figure which shows an example of the spectrum data which shows the result of having performed the instrumental analysis measurement with respect to the green tea sample by the instrumental analyzer shown in FIG. It is a figure which shows the spectrum data and the correlation which show the light absorbency of a certain green tea sample obtained in the quality prediction apparatus shown in FIG. It is a figure which shows the evaluation result of the prediction error in the PLS regression formula created using the PLS method in the 2nd to 4th derivative. It is a figure which shows the coefficient of the PLS regression formula in each effective wavelength in a 2nd-4th order differential coefficient. It is a figure which shows the correlation degree of the quality prediction value of the green tea which the model type | formula creation part shown in FIG. 1 shows, and the quality evaluation value of the green tea which is the result of the appraisal. It is a figure which shows the main screen started when the process in the quality prediction apparatus 1 and the equipment analyzer 2 shown in FIG. 1 is performed. It is a figure which shows an example of the model formula produced when the process in the quality prediction apparatus shown in FIG. 1 is performed with respect to the green tea sample.

Explanation of symbols

1 Quality prediction device 3 Main screen (operation screen)
11 Feature Extraction Unit (Feature Extraction Unit)
12 feature quantity evaluation value determination unit (feature quantity evaluation value determination means)
13 Correlation degree calculation unit (correlation degree calculation means)
14 Correlation degree determination unit (correlation degree determination means)
15 Effective wavelength determining unit (effective wavelength determining means)
16 Model formula creation part (quality predicted value determination means)
17 Display section (display means)

Claims (9)

A quality prediction device for predicting the quality of food,
A feature amount extraction means for extracting a feature amount of a spectrum for each food sample from instrument analysis results for a plurality of food samples;
Feature amount evaluation for classifying the spectrum feature amount extracted for each food sample by the feature amount extraction unit for each wavelength and determining a feature amount evaluation value for evaluating the ranking of the spectrum feature amount classified for each wavelength A value determining means;
Correlation degree calculating means for obtaining a correlation degree between the feature amount evaluation value determined by the feature amount evaluation value determining means and a quality evaluation value by an expert who performs quality evaluation of the plurality of food samples;
Correlation degree determination means for determining whether or not the absolute value of the correlation degree obtained by the correlation degree calculation means is larger than a preset correlation degree threshold;
When the correlation degree determining means determines that the absolute value of the correlation degree is larger than the threshold value, the wavelength in the feature value evaluation value used when the correlation degree calculating means obtains the correlation degree is set as an effective wavelength. An effective wavelength determining means for determining;
A quality prediction apparatus comprising: a quality prediction value determination unit that determines a quality prediction value of each food sample using the effective wavelength determined by the effective wavelength determination unit.   The feature amount extraction means obtains a higher-order differential coefficient for each wavelength by differentiating the device analysis result with respect to the wavelength, and extracts the same-order differential coefficient from the obtained differential coefficients to express the vector. The quality prediction apparatus according to claim 1, wherein a thing is extracted as a feature amount of the spectrum.   The quality prediction apparatus according to claim 1 or 2, wherein the quality prediction value determination means creates a model formula for determining a quality prediction value of each food sample by performing multivariate analysis.   The quality prediction apparatus according to claim 3, wherein the multivariate analysis is a regression analysis using a partial least square method.   The quality prediction apparatus according to any one of claims 1 to 4, wherein the instrumental analysis result is data indicating the absorbance of a food sample obtained by near infrared spectroscopy.   The quality prediction apparatus according to claim 1, further comprising display means for displaying an operation screen for acquiring a user operation. A quality prediction method in a quality prediction apparatus for predicting the quality of food,
A feature amount extraction step for extracting a spectrum feature amount for each food sample from instrument analysis results for a plurality of food samples;
Feature amount evaluation for categorizing the feature amount of the spectrum extracted for each food sample in the feature amount extraction step for each wavelength and determining a feature amount evaluation value for evaluating the rank of the feature amount of the spectrum classified for each wavelength A value determination step;
A degree-of-correlation calculation step for obtaining a degree of correlation between the feature amount evaluation value determined in the feature amount evaluation value determination step and a quality evaluation value by an expert who performs quality evaluation of the plurality of food samples;
A correlation degree determining step for determining whether or not the absolute value of the correlation degree obtained in the correlation degree calculating step is greater than a preset correlation degree threshold;
When it is determined in the correlation degree determination step that the absolute value of the correlation degree is larger than the threshold value, the wavelength in the feature amount evaluation value used when the correlation degree is obtained in the correlation degree calculation step is set as an effective wavelength. An effective wavelength determining step to determine;
A quality prediction method comprising: a quality prediction value determination step for determining a quality prediction value of each food sample using the effective wavelength determined in the effective wavelength determination step.   The quality prediction program for functioning a computer as said each means of the quality prediction apparatus of any one of Claim 1 to 6.   A computer-readable recording medium on which the quality prediction program according to claim 8 is recorded.

Images